Sustainability
We practise what
we preach
A platform built to improve energy efficiency should hold itself to the same standard. Every architectural decision, every line of code, every AI call — designed with carbon awareness.
This page visit
Estimated emissions per visit
Data transfer CO₂
Tree offset time
Powered by CO2.js / Sustainable Web Design model / Green hosted
Per AI analysis
Full property analysis (6 images)
Estimates based on GPT-5 inference costs, average image sizes of 2.5 MB, and Sustainable Web Design transfer model. GPU compute estimate derived from published ML energy research.
How an analysis
stays lean
AI is the most carbon-intensive part of what we do. Here's how we keep it deliberate.
Right-sized models
Individual image analysis uses GPT-5-mini. The full model only runs once for final synthesis. This cuts GPU compute by ~60% compared to running the full model on every image.
Results cached permanently
Every analysis is stored in Convex. Viewing your results a hundred times costs zero additional API calls. No TTL expiration — your data stays without re-computation.
User-triggered only
No speculative pre-analysis, no background polling, no auto-refresh. AI runs when you press the button. Every GPU cycle is intentional.
Parallel processing
All 6 images are analysed concurrently in a single batch, not sequentially. This reduces total wall-clock time and connection overhead significantly.
Efficient prompting
Structured output schemas mean the AI returns precisely what we need — no verbose prose to parse. Fewer output tokens means less compute per request.
Scoped outputs
Each analysis produces only the data needed for its context — component detection, EPC estimation, or report generation. No redundant re-analysis across stages.
The cost of understanding your home's energy performance
is roughly ~4.6g of CO₂
That's less than boiling a kettle. A full AI-powered property analysis — 6 images, component detection, EPC estimation — for the carbon cost of a few seconds of browsing.
Beyond the
analysis
Vercel edge network on renewable energy. Convex serverless backend that scales to zero. CDN-cached static assets cutting repeat transfers by 95%. Green hosting verified through The Green Web Foundation.
Next.js App Router with automatic code splitting. Tailwind CSS that purges unused styles. Dynamic imports for heavy libraries. Tree-shaking that eliminates dead code from production bundles.
WebP image optimisation via Next.js. Convex real-time sync sends only deltas, not full payloads. Incremental progress tracking — no re-uploading completed steps. Automatic cleanup of temporary analysis data.
Regular audits to remove unused packages. Preference for lightweight, well-maintained libraries. Zero-dependency utilities where a simple function replaces an entire npm package.
Core Web Vitals monitored via Vercel Analytics. Server Components render on the edge. Lazy loading for off-screen content. Responsive images sized per device to avoid oversized mobile downloads.
The bigger picture
UK homes produce 20% of the country's carbon emissions
Traditional EPC assessments are expensive and opaque. By making AI-powered energy analysis accessible to everyone, we lower the barrier to understanding — and acting on — the improvements that matter most. From insulation to heating systems, from draught proofing to smart controls.
Open source
Our carbon metrics use CO2.js from The Green Web Foundation. Transparent methodology means verifiable claims. The tools we rely on benefit the wider sustainability community.
Sustainability isn't a checkbox. This page is a living dashboard — as our stack evolves, so do these numbers.